DETAILED ACTION
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claim 19 is rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter.
Claim 19 recites “A computer program product comprising: a computer-readable storage media…”. As is readily known by the skilled artisan, any such “computer program product” and “computer-readable storage media”, given the broadest reasonable interpretation in the arts, necessarily include transmission-type media, such as signals and propagating waves, (since at least computer programs can produce signals carrying instructions thereon) non-statutory subject matter under 35 U.S.C. § 101.
Besides amending the claim by adding the terms “non-transitory” preceding the terms “computer program product” Examiner urges that a rejection under § 101 can also be avoided by either amending the claimed terms to: “computer usable memory,” “computer usable storage memory,” “computer readable memory,” “computer readable device,” “computer recordable memory,” “computer recordable device,” (i.e. any variations thereof, where “media” or “medium” is replaced by “device” or “memory”) or adding “wherein the media excludes signals”.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 4-12 and 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Liu et al. (US20250156687A1, also Provisional application No. 63/305,845, filed on Feb. 2, 2022) in view of Vaikuntanathan (US20200036512A1).
Regarding claim 1, Liu discloses a computer-implemented method for executing an artificial intelligence model, comprising (Fig 4 and para [0095] show a machine learning model may include neural network layers):
receiving an input for execution of the artificial intelligence model (para [0006] shows at least one first layer of the machine learning model configured to process a first input of data),
wherein the artificial intelligence model is split into an input block [first layer] (para [0006] shows at least one first layer of the machine learning model configured to process a first input of data; Fig 4 and para [0104] show computing device 104 may profile short-term data and may execute one or more first layers of the machine learning model to generate an output; para [0105] shows the output may be transmitted from the at least one computing device 104 to the server 102),
an intermediate block [server 102] (para [0006] shows generating, with the server, a classification based on the first input of data; Fig 4 and para [0107] show server 102 may execute the second portion of the machine learning model to finish the classification (e.g., inference) based on the short term and long-term data inputs), and
an output block [second layer] (para [0006] shows the second portion includes at least one second layer of the machine learning model configured to output the classification), such that
the input block receives specific input and provides an intermediate output [encrypted short-term data profile] (Fig 4 and para [0104] show computing device 104 may profile short-term (input) data to generate an output of the first portion of the machine learning model; para [0105] shows computing device 104 may use an encoder process to encode the output. The encrypted model data, which includes the output, may be transmitted from the at least one computing device 104 to the server 102),
the intermediate block receives as input the intermediate output [encrypted short-term data profile] and provides another intermediate output [short-term and long-term classification] (para [0106] shows server 102 may decode the encrypted model data received from the at least one computing device 104; para [0107] shows server 102 may finish the classification (e.g., inference) based on the short-term and long-term data inputs), and
the output block receives as input the another intermediate output [short-term and long-term classification] and provides specific output (para [0006] shows at least one second layer of the machine learning model configured to output the classification);
executing the input block [first layer] by a first computer system [computing device 104] using the input, producing a first intermediate output [short-term data profile]; encoding by the first computer system the first intermediate output using a first encoding protocol to produce an encoded first intermediate output [encrypted short-term data profile] (para [0006] shows at least one first layer of the machine learning model configured to process a first input of data; Fig 4 and para [0104] show computing device 104 may execute one or more first layers of the machine learning model to profile short-term (input) data to generate an output; para [0105] shows computing device 104 may use an encoder process to encode the output. The encrypted model data, which includes the output, may be transmitted from the at least one computing device 104 to the server 102);
sending the encoded first intermediate output [encrypted short-term data profile] to a second computer system [server 102] to allow the second computer system to (Fig 4 and para [0105] show the encrypted model data, which includes the output, may be transmitted from the at least one computing device 104 to the server 102):
decode the encoded first intermediate output [encrypted short-term data profile] using the first encoding protocol, and execute the intermediate block using as input the first intermediate output to produce a second intermediate output [short-term and long-term classification] (para [0106] shows server 102 may decode the encrypted model data received from the at least one computing device 104; para [0107] shows server 102 may finish the classification (e.g., inference) based on the short-term and long-term data inputs),
send the second intermediate output to the first computer system [computing device 104]; and executing the output block at the first computer system using as input the second intermediate output, producing a result output (para [0107] shows server 102 may finish the classification (e.g., inference) based on the short-term and long-term data inputs; para [0104] shows computing device 104 may receive a first portion of a machine learning model from the at least one server 102 and execute one or more first layers of the machine learning model to generate an output of the first portion of the machine learning model.)
Liu fails to show the second computer system [server 102] to encode the second intermediate output [short-term and long-term classification]. Specifically, Liu fails to show the second computer system to:
encode the second intermediate output using a second encoding protocol to produce an encoded second intermediate output, and
send the encoded second intermediate output to the first computer system; and
in response to receiving the encoded second intermediate output, decoding by the first computer system the encoded second intermediate output using the second encoding protocol.
However Vaikuntanathan, in an analogous art (para [0039] shows classification by a neural network), discloses the second computer system to (para [0012] shows the server comprising the trusted hardware):
encode the second intermediate output using a second encoding protocol to produce an encoded second intermediate output (para [0028] shows trusted hardware device 150 (e.g., server) may encrypt the result), and
send the encoded second intermediate output to the first computer system (para [0028] shows trusted hardware device 150 may transmit the encrypted result, e.g. via server 120, to the recipient device 160); and
in response to receiving the encoded second intermediate output, decoding by the first computer system the encoded second intermediate output using the second encoding protocol (para [0029] shows the recipient device 160 (e.g., client 110) may decrypt and merge the results.)
Computing device 104 and server 102 (Liu; Fig 4) are mapped to client 110 and server 120 (Vaikuntanathan; Fig 2), respectively. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify the method of Liu with the teaching of Vaikuntanathan in order for the client to merge the results of the computations by both the client and the server (Vaikuntanathan; para [0029]).
Regarding claim 4, Liu-Vaikuntanathan as applied to claim 1 discloses the splitting of the artificial intelligence model is performed based on available resources in the first computer system (Liu; para [0070] shows to improve the computational efficiency of a central server by splitting the execution of a machine learning model between at least one remote computing device (e.g., an edge device) and the central server; para [0072] shows the computing device 104 is remote (e.g., communicatively and physically arranged as an independent device) from the server 102.)
Regarding claim 5, Liu-Vaikuntanathan as applied to claim 4 discloses the splitting of the artificial intelligence model is dynamically performed or performed using one of predefined splitting options which are associated with a respective amount of resources (Liu; para [0070] shows to improve the computational efficiency of a central server by splitting the execution of a machine learning model between at least one remote computing device (e.g., an edge device) and the central server; para [0072] shows the computing device 104 is remote (e.g., communicatively and physically arranged as an independent device) from the server 102. Vaikuntanathan; para [0009] shows such a divide-and-conquer approach splits computations into different domains to optimize speed and security according to each type of data.)
Regarding claim 6, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1,
wherein the execution of the artificial intelligence model comprises execution of a succession of processing steps (Liu; Fig 4 shows steps 404-418),
wherein splitting the artificial intelligence model is performed such that the input block is configured to perform a first number of successive processing steps (Liu; para [0103] shows steps 402, 404, and 406 may be executed by the at least one computing device 104),
the intermediate block is configured to perform a second number of successive processing steps that follow the first number of successive processing steps of the input block (Liu; para [0103] shows steps 408, 410, 412, 414, 416, and 418 may be executed by the at least one server 102), and
the output block is configured to perform a third number of last successive processing steps (Liu; para [0006] shows at least one second layer of the machine learning model configured to output the classification),
wherein a sum of the first number of successive processing steps, the second number of successive processing steps, and the third number of last successive processing steps is a total number of processing steps in the artificial intelligence model (Liu; Fig 4 shows steps 404-418).
Regarding claim 7, Liu-Vaikuntanathan discloses the computer-implemented method of claim 6,
wherein the first number of successive processing steps is smaller than the second number of successive processing steps by a first delta value (Liu; para [0103] shows steps 402, 404, and 406 may be executed by the at least one computing device 104; para [0103] shows steps 408, 410, 412, 414, 416, and 418 may be executed by the at least one server 102),
wherein the third number of last successive processing steps is smaller than the second number of successive processing steps by a second delta value (Liu; para [0103] shows steps 408, 410, 412, 414, 416, and 418 may be executed by the at least one server 102); para [0006] shows at least one second layer of the machine learning model configured to output the classification).
Regarding claim 8, Liu-Vaikuntanathan discloses the computer-implemented method of claim 7, further comprising determining the first and second delta values based on available resources in the first computer system (Liu; para [0070] shows to improve the computational efficiency of a central server by splitting the execution of a machine learning model between at least one remote computing device (e.g., an edge device) and the central server. In this manner, the central server will require a fraction of the computational resources; para [0072] shows the computing device 104 is remote (e.g., communicatively and physically arranged as an independent device) from the server 102. Vaikuntanathan; para [0009] shows such a divide-and-conquer approach splits computations into different domains to optimize speed and security according to each type of data.)
Regarding claim 9, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the first encoding protocol is the second encoding protocol (Liu; para [0105] shows the at least one computing device 104 may use an encoder process to encode the output of the first portion of the machine learning model. Encoding may include encryption of the output data. Vaikuntanathan; para [0012] shows the server comprising the trusted hardware; para [0028] shows trusted hardware device 150 (e.g., server) may encrypt the result.)
Regarding claim 10, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the first encoding protocol is different from the second encoding protocol (Liu; para [0105] shows the at least one computing device 104 may use an encoder process to encode the output of the first portion of the machine learning model. Encoding may include encryption and compression of the output data. Vaikuntanathan; para [0012] shows the server comprising the trusted hardware; para [0028] shows trusted hardware device 150 (e.g., server) may encrypt the result.)
Regarding claim 11, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the first encoding protocol is selected from a group consisting of compression and encryption, and wherein the second encoding protocol is selected from a group consisting of compression and encryption (Liu; para [0105] shows encoding may include compression of the output data.)
Regarding claim 12, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the execution of the artificial intelligence model is an inference of the artificial intelligence model which is already trained (Liu; para [0078] shows a model trained on historic data; para [0107] shows classification (e.g., inference) based on the short term and long-term data inputs.)
Regarding claim 14, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the first computer system has an amount of processing resources which is smaller than the processing resources of the second computer system (Liu; para [0060] shows a mobile device; para [0070] shows a central server.)
Regarding claim 15, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the first computer system is selected from a group consisting of an edge device and an internet of things (IoT) device (Liu; para [0061] shows a mobile device executing an electronic wallet application; para [0070] shows at least one remote computing device (e.g., an edge device).)
Regarding claim 16, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the second computer system is provided as a service in a cloud environment (Liu; para [0073] shows a cloud computing network. Vaikuntanathan; para [0012] shows the server is a Cloud service comprising the trusted hardware.)
Regarding claim 17, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, wherein the artificial intelligence model is a foundation model, and wherein the artificial intelligence model is a deep neural network where the input block represents first network layers, the intermediate block represents middle network layers, and the output block represents last network layers (Liu; para [0078] shows a model trained on historic data; para [0095] shows the layers may be fully connected neural network layers. The layers may include one or more input layers, hidden layers, and output layers.)
Regarding claim 18, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, further comprising splitting the artificial intelligence model by a management server, and deploying by the management server the input block, the output block, and the intermediate block in the first and second computer systems (Liu; para [0065] shows the services may be managed by a service provider).
Regarding claim 19, claim 19 is directed to a computer program product. Claim 19 requires limitations that are similar to those recited in the method claim 1 to carry out the method steps. And since the references of Liu-Vaikuntanathan combined teach the method including limitations required to carry out the method steps, therefore claim 19 would have also been obvious in view of the method disclosed in Liu-Vaikuntanathan combined.
Regarding claim 20, claim 20 is directed to a system. Claim 20 requires limitations that are similar to those recited in the method claim 1 to carry out the method steps. And since the references of Liu-Vaikuntanathan combined teach the method including limitations required to carry out the method steps, therefore claim 20 would have also been obvious in view of the method disclosed in Liu-Vaikuntanathan combined.
Furthermore, Liu-Vaikuntanathan as combined discloses one or more computer readable storage media storing program instructions and one or more processors (Liu; para [0084]).
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Vaikuntanathan, further in view of Ananthanarayanan et al. (US20220383188A1).
Regarding claim 2, Liu-Vaikuntanathan discloses the computer-implemented method of claim 1, further comprising: before executing the output block, deleting the input block from the first computer system.
However, Ananthanarayanan discloses before executing the output block, deleting the input block from the first computer system (para [0023] shows the network edge using machine learning technologies, such as neural networks, to train analytics models; para [0033] shows the model merger 204 merges models for reducing the memory footprint of models in the edge; para [0034] shows the model merger 204 merges selected layers of different models; para [0036] shows two layers are sharable when both layers have the same input data size. By consolidating (e.g., deleting) sharable layers that are high in memory consumption, the technology may efficiently reduce a memory footprint of the overall models (e.g., by merging the heaviest layers first).)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify the method of Liu-Vaikuntanathan with the teaching of Ananthanarayanan in order to reduce the memory footprint of models in the edge (Ananthanarayanan; para [0033]).
Regarding claim 3, Liu-Vaikuntanathan as applied to claim 1 fails to teach after executing the input block, deploying the output block into the first computer system.
However, Ananthanarayanan discloses after executing the input block, deploying the output block into the first computer system (para [0023] shows the network edge using machine learning technologies, such as neural networks, to train analytics models; para [0033] shows the model merger 204 merges models for reducing the memory footprint of models in the edge; para [0034] shows the model merger 204 merges selected layers of different models; para [0060] shows following the transmit operation 616 that transmits (e.g., deploys) the merged model, receive operation 652 receives information associated with data drift as detected by output of the merged model).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify the method of Liu-Vaikuntanathan with the teaching of Ananthanarayanan in order to to reduce the memory footprint of models in the edge (Ananthanarayanan; para [0033]).
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Liu in view of Vaikuntanathan, further in view of Ben-Itzhak et al. (US20220292342A1).
Regarding claim 13, Liu-Vaikuntanathan as applied to claim 1 fails to teach:
for each further received input, training the artificial intelligence model, wherein the first computer system is further configured to compute in each iteration a loss function and to send a result to the second computer system, wherein the result is used by the first and second computer systems to update learnable parameters of the artificial intelligence model, wherein an iteration is performed until the loss function fulfils a convergence criterion.
However, Ben-Itzhak discloses the first computer system is further configured to compute in each iteration a loss function and to send a result to the second computer system, wherein the result is used by the first and second computer systems to update learnable parameters of the artificial intelligence model, wherein an iteration is performed until the loss function fulfils a convergence criterion (para [0001] shows multiple distributed clients - under the direction of a central server known as a parameter server – collaboratively train an machine learning (ML) model; where the ML model is a neural network, learning typically proceeds as follows: (1) the parameter server transmits a copy of the neural network comprising the neural network's current parameter values to a subset of the clients; (2) each client in the subset provides one or more data instances in its local training dataset as input to its received neural network (resulting in one or more corresponding results/predictions), computes a gradient for the neural network (referred to herein as a “client gradient”) based on the inputted data instances and outputted results/predictions according to a loss function, and transmits the client gradient to the parameter server; (3) the parameter server aggregates the client gradients into a global gradient and uses the global gradient to update the parameters of the neural network (i.e., performs an optimization step); and (4) steps (1) through (3) are repeated until a termination criterion is met.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to modify the method of Liu-Vaikuntanathan with the teaching of Ben-Itzhak in order to repeat an optimization step until a termination criterion is met (Ben-Itzhak; para [0001]).
Citation of Relevant Prior Art
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Vald et al. (US20200382273A1) shows in para [0021] the machine learning module 132 may be used to train neural networks; para [0031] shows in the case of a request from machine learning device 130 to re-encrypt encrypted data 146, the privacy preserving server 150 encryptor 180 to produce encrypted output 182 and return encrypted output to machine learning device 130; para [0033] shows the machine learning device obtains encrypted data from a data lake. This data lake allows access to the encrypted data to all devices within the data lake, but only provides encrypted data to the devices and may not provide means to decrypt the encrypted data to the devices. Thus, any operation involving decrypting the encrypted data also involves use of an privacy preserving server associated with the data lake.
Gharibi et al. (US20240154942A1) discloses in para [0048] a model (neural network) is split into two parts: one part (A) resides on the client side and includes the input layer, and the other part (B) resides on the server side and often includes the output layer; para [0042] shows in each iteration, the server 102 averages all participating models to create a trained model B; para [0045] shows the server 202 calculates the loss function and the server 202 does backpropagation and calculates gradients at the S layer. The server 202 sends the gradients of S only (i.e., SB (206C), SB2 (208C), SBN (210C)) to the respective client 206, 208, 210.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAN DOAN whose telephone number is (571)270-0162. The examiner can normally be reached Monday - Friday 8am - 5pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Oscar Louie, can be reached at (571) 270-1684. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/TAN DOAN/Primary Examiner, Art Unit 2445